Identify a live route by prefix and peer (closes #55)

A peer has at most one route for a prefix, but the live route tables were
keyed by prefix, origin AS and peer. An announcement with a new origin AS
added a second row instead of replacing the first, and a withdrawal sent in
the same message as announcements was given their origin AS and so missed
the withdrawn route.

The tables are now unique on prefix and peer. An announcement updates the
peer's row for the prefix, origin AS included, and a withdrawal deletes it
without looking at an origin AS. The origin-only delete statements, the
deletion's origin field and DeleteLiveRoute's origin parameter are removed.

Model: opus-5-5

README.md

@@ -144,8 +144,9 @@ routewatch/
 - **Reconnection**: Exponential backoff (5s-320s) with reset after 30s of stable connection
 - **IP Lookup**: the most specific live route is found by looking up the
   address's prefix at each mask length, longest first, in the prefix index
-  (at most 33 lookups for IPv4, 129 for IPv6); prefixes are stored in the
+  (at most 25 lookups for IPv4, 113 for IPv6); prefixes are stored in the
-  text form Go's `net/netip` prints
+  text form Go's `net/netip` prints. Routes shorter than /8 for IPv4 or /16
+  for IPv6, such as a default route, are not used.
 
 ### Database Schema
 

internal/database/counts_test.go

@@ -128,7 +128,7 @@ func TestLiveCountsTrackWritesInRealtime(t *testing.T) {
 
 	// A withdrawal removes one route.
 	if err := db.DeleteLiveRouteBatch([]LiveRouteDeletion{
-		{Prefix: "203.0.113.0/24", OriginASN: 64501, PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
+		{Prefix: "203.0.113.0/24", PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
 	}); err != nil {
 		t.Fatalf("DeleteLiveRouteBatch: %v", err)
 	}
@@ -327,13 +327,11 @@ func announceRoutes(db *Database, batch bool, routes []*LiveRoute) error {
 }
 
 // withdrawRoutes removes routes in one DeleteLiveRouteBatch, or with one
-// DeleteLiveRoute each. It names each route by prefix and peer only, with no
+// DeleteLiveRoute each.
-// origin ASN, as a withdrawal from the feed does when its message carries no AS
-// path.
 func withdrawRoutes(db *Database, batch bool, routes []*LiveRoute) error {
 	if !batch {
 		for _, route := range routes {
-			if err := db.DeleteLiveRoute(route.Prefix, 0, route.PeerIP); err != nil {
+			if err := db.DeleteLiveRoute(route.Prefix, route.PeerIP); err != nil {
 				return err
 			}
 		}

internal/database/database.go

@@ -37,6 +37,14 @@ const (
 	ipv6Bits       = 128
 )
 
+// The IP lookup does not use routes shorter than these mask lengths. A default
+// route or a route this short only says where a peer sends traffic it has no
+// better route for; it does not say who holds the address.
+const (
+	minLookupMaskLengthV4 = 8
+	minLookupMaskLengthV6 = 16
+)
+
 // SQLite memory tuning. cache_size and busy_timeout go in the DSN so every
 // pooled connection gets them; the heap limits are process-wide and set once.
 const (
@@ -220,21 +228,22 @@ func (d *Database) beginTx() (*loggingTx, error) {
 }
 
 // A live-route upsert is an UPDATE followed, only when no row matched, by an
-// INSERT. The UPDATE's rows-affected count (1 for an existing key, 0 for a new
+// INSERT. A live route is keyed by prefix and peer, so an announcement replaces
-// one) is what lets the in-memory route counters stay exact without a COUNT(*).
+// the peer's route for the prefix whatever its origin AS. The UPDATE's
-// Callers hold the database write lock, so no other writer can insert the same
+// rows-affected count (1 for an existing key, 0 for a new one) is what lets the
-// key between the two statements. The id column is set only on INSERT, so an
+// in-memory route counters stay exact without a COUNT(*). Callers hold the
-// updated route keeps its original id, exactly as the previous ON CONFLICT
+// database write lock, so no other writer can insert the same key between the
-// upsert did.
+// two statements. The id column is set only on INSERT, so an updated route
+// keeps its original id, exactly as the previous ON CONFLICT upsert did.
 const (
-	updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, as_path = ?, next_hop = ?,
+	updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, origin_asn = ?, as_path = ?,
-		last_updated = ? WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`
+		next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
 
 	insertLiveRouteV4SQL = `INSERT INTO live_routes_v4 (id, prefix, mask_length, origin_asn, peer_ip,
 		as_path, next_hop, last_updated) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`
 
-	updateLiveRouteV6SQL = `UPDATE live_routes_v6 SET mask_length = ?, as_path = ?, next_hop = ?,
+	updateLiveRouteV6SQL = `UPDATE live_routes_v6 SET mask_length = ?, origin_asn = ?, as_path = ?,
-		last_updated = ? WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`
+		next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
 
 	insertLiveRouteV6SQL = `INSERT INTO live_routes_v6 (id, prefix, mask_length, origin_asn, peer_ip,
 		as_path, next_hop, last_updated) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`
@@ -253,8 +262,8 @@ const (
 // live route for its prefix.
 func upsertRouteRowV4(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string) (
 	inserted, newPrefix bool, err error) {
-	res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
+	res, err := upd.Exec(route.MaskLength, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
-		route.Prefix, route.OriginASN, route.PeerIP)
+		route.Prefix, route.PeerIP)
 	if err != nil {
 		return false, false, err
 	}
@@ -285,8 +294,8 @@ func upsertRouteRowV4(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string
 // live route for its prefix.
 func upsertRouteRowV6(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string) (
 	inserted, newPrefix bool, err error) {
-	res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
+	res, err := upd.Exec(route.MaskLength, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
-		route.Prefix, route.OriginASN, route.PeerIP)
+		route.Prefix, route.PeerIP)
 	if err != nil {
 		return false, false, err
 	}
@@ -433,32 +442,18 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
 		}
 	}()
 
-	// No longer need to separate deletions since we handle them in the loop below
-
 	// Prepare statements for both IPv4 and IPv6 tables
-	stmtV4WithOrigin, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`)
+	deleteV4, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`)
 	if err != nil {
-		return fmt.Errorf("failed to prepare IPv4 delete with origin statement: %w", err)
+		return fmt.Errorf("failed to prepare IPv4 delete statement: %w", err)
 	}
-	defer func() { _ = stmtV4WithOrigin.Close() }()
+	defer func() { _ = deleteV4.Close() }()
 
-	stmtV4WithoutOrigin, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`)
+	deleteV6, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`)
 	if err != nil {
-		return fmt.Errorf("failed to prepare IPv4 delete without origin statement: %w", err)
+		return fmt.Errorf("failed to prepare IPv6 delete statement: %w", err)
 	}
-	defer func() { _ = stmtV4WithoutOrigin.Close() }()
+	defer func() { _ = deleteV6.Close() }()
-
-	stmtV6WithOrigin, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`)
-	if err != nil {
-		return fmt.Errorf("failed to prepare IPv6 delete with origin statement: %w", err)
-	}
-	defer func() { _ = stmtV6WithOrigin.Close() }()
-
-	stmtV6WithoutOrigin, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`)
-	if err != nil {
-		return fmt.Errorf("failed to prepare IPv6 delete without origin statement: %w", err)
-	}
-	defer func() { _ = stmtV6WithoutOrigin.Close() }()
 
 	hasV4, err := tx.Prepare(prefixHasLiveRouteV4SQL)
 	if err != nil {
@@ -477,38 +472,18 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
 	// Mask lengths of the prefixes this batch leaves with no live route.
 	var gonePrefixMaskLengthsV4, gonePrefixMaskLengthsV6 []int
 	for _, del := range deletions {
-		var stmt *sql.Stmt
+		stmt := deleteV4
-
+		if del.IPVersion != ipVersionV4 {
-		// Select appropriate statement based on IP version and whether we have origin ASN
+			stmt = deleteV6
-		//nolint:nestif // Clear logic for selecting the right statement
-		if del.IPVersion == ipVersionV4 {
-			if del.OriginASN == 0 {
-				stmt = stmtV4WithoutOrigin
-			} else {
-				stmt = stmtV4WithOrigin
-			}
-		} else {
-			if del.OriginASN == 0 {
-				stmt = stmtV6WithoutOrigin
-			} else {
-				stmt = stmtV6WithOrigin
-			}
-		}
-
-		// Execute deletion
-		var res sql.Result
-		if del.OriginASN == 0 {
-			res, err = stmt.Exec(del.Prefix, del.PeerIP)
-		} else {
-			res, err = stmt.Exec(del.Prefix, del.OriginASN, del.PeerIP)
 		}
 
+		res, err := stmt.Exec(del.Prefix, del.PeerIP)
 		if err != nil {
 			return fmt.Errorf("failed to delete route %s: %w", del.Prefix, err)
 		}
 
-		// A deletion with no origin ASN can remove several rows, so use the
+		// The feed often withdraws a route that is not live, which deletes
-		// exact rows-affected count to keep the in-memory route counters right.
+		// nothing, so the in-memory route counters use the rows-affected count.
 		affected, err := res.RowsAffected()
 		if err != nil {
 			return fmt.Errorf("failed to count deleted route %s: %w", del.Prefix, err)
@@ -1274,9 +1249,8 @@ func (d *Database) UpsertLiveRoute(route *LiveRoute) error {
 	return nil
 }
 
-// DeleteLiveRoute deletes a live route
+// DeleteLiveRoute deletes the live route the peer has for the prefix
-// If originASN is 0, deletes all routes for the prefix/peer combination
+func (d *Database) DeleteLiveRoute(prefix, peerIP string) error {
-func (d *Database) DeleteLiveRoute(prefix string, originASN int, peerIP string) error {
 	d.lock("DeleteLiveRoute")
 	defer d.unlock()
 
@@ -1301,22 +1275,12 @@ func (d *Database) DeleteLiveRoute(prefix string, originASN int, peerIP string)
 	}()
 
 	// Literal per-table queries (rather than one formatted with the table name)
-	// so the delete carries no dynamically built SQL. A delete with no origin
+	// so the delete carries no dynamically built SQL.
-	// ASN can remove several rows.
 	var res sql.Result
-	switch {
+	if isV4 {
-	case isV4 && originASN == 0:
 		res, err = tx.Exec(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`, prefix, peerIP)
-	case isV4:
+	} else {
-		res, err = tx.Exec(
-			`DELETE FROM live_routes_v4 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`,
-			prefix, originASN, peerIP)
-	case originASN == 0:
 		res, err = tx.Exec(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`, prefix, peerIP)
-	default:
-		res, err = tx.Exec(
-			`DELETE FROM live_routes_v6 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`,
-			prefix, originASN, peerIP)
 	}
 	if err != nil {
 		return err
@@ -1876,11 +1840,12 @@ func (d *Database) GetIPInfo(ip string) (*IPInfo, error) {
 // GetIPInfoContext returns comprehensive IP information with context support.
 //
 // It finds the most specific live route for the address with point lookups on
-// the prefix index: for each mask length from the longest (32 or 128) down to 0
+// the prefix index: for each mask length from the longest (32 or 128) down to
-// it looks up the address's network prefix at that length, and the first length
+// minLookupMaskLengthV4 or minLookupMaskLengthV6 it looks up the address's
-// with a live route wins. That is at most 33 or 129 indexed lookups, however
+// network prefix at that length, and the first length with a live route wins.
-// many routes are live. The prefix is looked up in the text form net/netip
+// That is at most 25 or 113 indexed lookups, however many routes are live. The
-// prints, which is the form the prefix handler stores.
+// prefix is looked up in the text form net/netip prints, which is the form the
+// prefix handler stores.
 func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, error) {
 	addr, err := netip.ParseAddr(ip)
 	if err != nil {
@@ -1891,9 +1856,11 @@ func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, er
 
 	info := &IPInfo{IP: ip, IPVersion: ipVersionV6}
 	routesTable, prefixesTable := "live_routes_v6", "prefixes_v6"
+	minMaskLength := minLookupMaskLengthV6
 	if addr.Is4() {
 		info.IPVersion = ipVersionV4
 		routesTable, prefixesTable = "live_routes_v4", "prefixes_v4"
+		minMaskLength = minLookupMaskLengthV4
 	}
 
 	// Route info with peer count and prefix first_seen
@@ -1934,7 +1901,7 @@ func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, er
 	var whoisUpdatedAt sql.NullTime
 
 	found := false
-	for bits := addr.BitLen(); bits >= 0; bits-- {
+	for bits := addr.BitLen(); bits >= minMaskLength; bits-- {
 		prefix := netip.PrefixFrom(addr, bits).Masked().String()
 		err := stmt.QueryRowContext(ctx, prefix).Scan(
 			&info.Netblock,

internal/database/database_test.go

@@ -13,6 +13,7 @@ import (
 	"git.eeqj.de/sneak/routewatch/internal/config"
 	"git.eeqj.de/sneak/routewatch/internal/logger"
 	"github.com/google/uuid"
+	"github.com/mattn/go-sqlite3"
 )
 
 // tempStoreMemory is the PRAGMA temp_store value meaning "hold temp B-trees in
@@ -122,8 +123,8 @@ func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
 	lookup("2001:db9::1", "")
 
 	err = db.DeleteLiveRouteBatch([]LiveRouteDeletion{
-		{Prefix: "10.1.2.0/24", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV4},
+		{Prefix: "10.1.2.0/24", PeerIP: testPeerIP, IPVersion: ipVersionV4},
-		{Prefix: "2001:db8:1:2::/64", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV6},
+		{Prefix: "2001:db8:1:2::/64", PeerIP: testPeerIP, IPVersion: ipVersionV6},
 	})
 	if err != nil {
 		t.Fatalf("DeleteLiveRouteBatch: %v", err)
@@ -133,6 +134,112 @@ func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
 	lookup("2001:db8:1:2::3", "2001:db8:1::/48")
 }
 
+// TestGetIPInfoIgnoresShortRoutes checks that the IP lookup does not report a
+// default route, or a route shorter than /8 for IPv4 or /16 for IPv6, as the
+// route for an address, and that a route of exactly /8 or /16, or longer, still
+// is. The prefixes are written out so that a wrong limit fails the test.
+func TestGetIPInfoIgnoresShortRoutes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(prefix string, asn int) {
+		t.Helper()
+
+		err := db.UpsertLiveRouteBatch([]*LiveRoute{{
+			ID:          uuid.New(),
+			Prefix:      prefix,
+			MaskLength:  netip.MustParsePrefix(prefix).Bits(),
+			IPVersion:   detectIPVersion(prefix),
+			OriginASN:   asn,
+			PeerIP:      testPeerIP,
+			ASPath:      []int{asn},
+			NextHop:     testPeerIP,
+			LastUpdated: ts,
+		}})
+		if err != nil {
+			t.Fatalf("UpsertLiveRouteBatch(%s): %v", prefix, err)
+		}
+	}
+
+	// lookup checks that ip resolves to the live prefix netblock from asn.
+	lookup := func(ip, netblock string, asn int) {
+		t.Helper()
+
+		info, err := db.GetIPInfo(ip)
+		if err != nil {
+			t.Errorf("GetIPInfo(%s): %v", ip, err)
+
+			return
+		}
+		if info.Netblock != netblock || info.ASN != asn {
+			t.Errorf("GetIPInfo(%s) = %s AS%d, want %s AS%d", ip, info.Netblock, info.ASN, netblock, asn)
+		}
+	}
+
+	// Default and very short routes, and routes one bit shorter than each limit.
+	announce("2000::/3", 64500)
+	announce("0.0.0.0/0", 64500)
+	announce("2000::/15", 64500)
+	announce("192.0.0.0/7", 64500)
+
+	for _, ip := range []string{"2001:db8::1", "192.0.2.1"} {
+		if info, err := db.GetIPInfo(ip); !errors.Is(err, ErrNoRoute) {
+			t.Errorf("GetIPInfo(%s) with only short routes = %+v, %v; want ErrNoRoute", ip, info, err)
+		}
+	}
+
+	// Routes of exactly the minimum length are used.
+	announce("2001::/16", 64501)
+	announce("192.0.0.0/8", 64501)
+
+	lookup("2001:db8::1", "2001::/16", 64501)
+	lookup("192.0.2.1", "192.0.0.0/8", 64501)
+
+	announce("2001:db8::/32", 64502)
+
+	lookup("2001:db8::1", "2001:db8::/32", 64502)
+}
+
+// TestLiveRouteTablesHoldOneRoutePerPrefixAndPeer checks that each live route
+// table refuses a second route from a peer for a prefix, even with another
+// origin AS. An upsert updates the peer's route before it would insert one, so
+// only an insert straight into the table shows what the table allows.
+func TestLiveRouteTablesHoldOneRoutePerPrefixAndPeer(t *testing.T) {
+	db, err := New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	const otherOriginASN = 64501
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	for _, tt := range []struct {
+		route     *LiveRoute
+		insertSQL string
+	}{
+		{route: mkV4Route(t, "198.51.100.0/24", 64500, ts), insertSQL: insertLiveRouteV4SQL},
+		{route: mkV6Route("2001:db8::/32", 64500, ts), insertSQL: insertLiveRouteV6SQL},
+	} {
+		if err := db.UpsertLiveRoute(tt.route); err != nil {
+			t.Fatalf("UpsertLiveRoute(%s): %v", tt.route.Prefix, err)
+		}
+
+		_, err := db.db.Exec(tt.insertSQL, uuid.NewString(), tt.route.Prefix, tt.route.MaskLength,
+			otherOriginASN, tt.route.PeerIP, "[]", tt.route.NextHop, ts)
+		var sqliteErr sqlite3.Error
+		if !errors.As(err, &sqliteErr) || sqliteErr.ExtendedCode != sqlite3.ErrConstraintUnique {
+			t.Errorf("second route from peer %s for %s: error %v, want a UNIQUE constraint error",
+				tt.route.PeerIP, tt.route.Prefix, err)
+		}
+	}
+}
+
 // TestConnectionPoolPragmas holds several pooled connections open at once and
 // checks each one carries the per-connection settings from the DSN, plus the
 // process-wide hard heap limit, and sees the new file with auto_vacuum
@@ -234,7 +341,6 @@ func TestVacuumReturnsFreePages(t *testing.T) {
 		routes = append(routes, route)
 		deletions = append(deletions, LiveRouteDeletion{
 			Prefix:    route.Prefix,
-			OriginASN: asn,
 			PeerIP:    route.PeerIP,
 			IPVersion: ipVersionV4,
 		})

internal/database/interface.go

@@ -55,7 +55,7 @@ type Store interface {
 	// Live route operations
 	UpsertLiveRoute(route *LiveRoute) error
 	UpsertLiveRouteBatch(routes []*LiveRoute) error
-	DeleteLiveRoute(prefix string, originASN int, peerIP string) error
+	DeleteLiveRoute(prefix, peerIP string) error
 	DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error
 	GetPrefixDistribution() (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
 	GetPrefixDistributionContext(ctx context.Context) (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)

internal/database/models.go

@@ -114,7 +114,6 @@ type IPInfo struct {
 // LiveRouteDeletion represents parameters for deleting a live route
 type LiveRouteDeletion struct {
 	Prefix    string
-	OriginASN int
 	PeerIP    string
 	IPVersion int
 }

internal/database/schema.sql

@@ -107,7 +107,7 @@ CREATE TABLE IF NOT EXISTS live_routes_v4 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	UNIQUE(prefix, origin_asn, peer_ip)
+	UNIQUE(prefix, peer_ip) -- a peer has at most one route for a prefix
 );
 
 -- IPv6 routing table maintained by PrefixHandler
@@ -120,7 +120,7 @@ CREATE TABLE IF NOT EXISTS live_routes_v6 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	UNIQUE(prefix, origin_asn, peer_ip)
+	UNIQUE(prefix, peer_ip) -- a peer has at most one route for a prefix
 );
 
 -- Indexes for live_routes_v4 table

internal/metrics/metrics.go

@@ -60,17 +60,22 @@ func New() *Tracker {
 	}
 }
 
-// SetConnected updates the connection status
+// SetConnected updates the connection status. Every connection after the
+// first counts one reconnect.
 func (t *Tracker) SetConnected(connected bool) {
-	wasConnected := t.isConnected.Swap(connected)
+	t.isConnected.Store(connected)
-	if connected {
+	if !connected {
-		t.mu.Lock()
+		return
-		t.connectedSince = time.Now()
+	}
-		t.mu.Unlock()
+
-		// Increment reconnect count (but not for the initial connection)
+	t.mu.Lock()
-		if wasConnected || t.reconnectCount.Load() > 0 {
+	// connectedSince is zero until the first connection.
-			t.reconnectCount.Add(1)
+	connectedBefore := !t.connectedSince.IsZero()
-		}
+	t.connectedSince = time.Now()
+	t.mu.Unlock()
+
+	if connectedBefore {
+		t.reconnectCount.Add(1)
 	}
 }
 

internal/metrics/metrics_test.go

@@ -0,0 +1,21 @@
+package metrics
+
+import "testing"
+
+// TestReconnectCount connects three times with a disconnect between each, the
+// way the streamer does, and expects every connection after the first to count
+// one reconnect.
+func TestReconnectCount(t *testing.T) {
+	tracker := New()
+
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+
+	const wantReconnects = 2
+	if got := tracker.GetReconnectCount(); got != wantReconnects {
+		t.Fatalf("reconnect count = %d, want %d", got, wantReconnects)
+	}
+}

internal/routewatch/app_integration_test.go

@@ -205,7 +205,7 @@ func (m *mockStore) UpsertLiveRoute(route *database.LiveRoute) error {
 }
 
 // DeleteLiveRoute mock implementation
-func (m *mockStore) DeleteLiveRoute(prefix string, originASN int, peerIP string) error {
+func (m *mockStore) DeleteLiveRoute(prefix, peerIP string) error {
 	// Simple mock - just return nil
 	return nil
 }

internal/routewatch/cli.go

@@ -20,6 +20,8 @@ const (
 	debugInterval = 60 * time.Second
 	// bytesPerMB is bytes per megabyte
 	bytesPerMB = 1024 * 1024
+	// failureExitCode is the exit status when RouteWatch stops on an error
+	failureExitCode = 1
 )
 
 // logDebugStats logs goroutine count and memory usage
@@ -83,6 +85,9 @@ func CLIEntry() {
 					go func() {
 						if err := rw.Run(ctx); err != nil {
 							logger.Error("RouteWatch error", "error", err)
+							if err := shutdowner.Shutdown(fx.ExitCode(failureExitCode)); err != nil {
+								logger.Error("Failed to shutdown gracefully", "error", err)
+							}
 						}
 					}()
 

internal/routewatch/prefixhandler.go

@@ -105,7 +105,24 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
-	// Process announcements
+	// Withdrawals are queued before announcements, because the batch keeps
+	// each peer's last update for a prefix: a message that lists a prefix as
+	// both withdrawn and announced keeps the route (RFC 4271 section 4.3).
+	// A withdrawal names only the prefix and peer: the message's AS path, and
+	// so its origin AS, belongs to the announced prefixes.
+	for _, prefix := range msg.Withdrawals {
+		h.batch = append(h.batch, prefixUpdate{
+			prefix:      canonicalPrefix(prefix),
+			peer:        msg.Peer,
+			messageType: "withdrawal",
+			timestamp:   timestamp,
+		})
+		// Record withdrawal in metrics
+		if h.metrics != nil {
+			h.metrics.RecordWithdrawal()
+		}
+	}
+
 	for _, announcement := range msg.Announcements {
 		for _, prefix := range announcement.Prefixes {
 			h.batch = append(h.batch, prefixUpdate{
@@ -123,22 +140,6 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 		}
 	}
 
-	// Process withdrawals
-	for _, prefix := range msg.Withdrawals {
-		h.batch = append(h.batch, prefixUpdate{
-			prefix:      canonicalPrefix(prefix),
-			originASN:   originASN, // Use the originASN from path if available
-			peer:        msg.Peer,
-			messageType: "withdrawal",
-			timestamp:   timestamp,
-			path:        msg.Path,
-		})
-		// Record withdrawal in metrics
-		if h.metrics != nil {
-			h.metrics.RecordWithdrawal()
-		}
-	}
-
 	// Check if we need to flush
 	if len(h.batch) >= prefixBatchSize {
 		h.flushBatchLocked()
@@ -179,14 +180,14 @@ func (h *PrefixHandler) flushBatchLocked() {
 	startTime := time.Now()
 	batchSize := len(h.batch)
 
-	// Group updates by prefix to deduplicate
+	// For each prefix and peer, keep only the last update the feed delivered.
-	// For each prefix, keep the latest update
+	// Updates from different peers for the same prefix are all kept. Arrival
-	prefixMap := make(map[string]prefixUpdate)
+	// order decides, not the timestamp: timestamps are whole seconds, so
+	// updates within one second carry the same one.
+	type prefixAndPeer struct{ prefix, peer string }
+	lastUpdates := make(map[prefixAndPeer]prefixUpdate)
 	for _, update := range h.batch {
-		key := update.prefix
+		lastUpdates[prefixAndPeer{update.prefix, update.peer}] = update
-		if existing, ok := prefixMap[key]; !ok || update.timestamp.After(existing.timestamp) {
-			prefixMap[key] = update
-		}
 	}
 
 	// Collect routes to upsert and delete
@@ -196,7 +197,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	// Collect unique prefixes to update
 	prefixesToUpdate := make(map[string]time.Time)
 
-	for _, update := range prefixMap {
+	for _, update := range lastUpdates {
 		// Track prefix for both announcements and withdrawals
 		if _, exists := prefixesToUpdate[update.prefix]; !exists || update.timestamp.After(prefixesToUpdate[update.prefix]) {
 			prefixesToUpdate[update.prefix] = update.timestamp
@@ -220,7 +221,6 @@ func (h *PrefixHandler) flushBatchLocked() {
 			// Create deletion record for batch delete
 			routesToDelete = append(routesToDelete, database.LiveRouteDeletion{
 				Prefix:    update.prefix,
-				OriginASN: update.originASN,
 				PeerIP:    update.peer,
 				IPVersion: ipVersion,
 			})
@@ -255,7 +255,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	elapsed := time.Since(startTime)
 	h.logger.Debug("Flushed prefix batch",
 		"batch_size", batchSize,
-		"unique_prefixes", len(prefixMap),
+		"unique_prefixes", len(prefixesToUpdate),
 		"success", successCount,
 		"duration_ms", elapsed.Milliseconds(),
 	)
@@ -424,51 +424,24 @@ func (h *PrefixHandler) processAnnouncementDirect(update prefixUpdate) {
 // processWithdrawalDirect handles removing a route directly without prefix table
 // nolint:unused // kept for potential future use
 func (h *PrefixHandler) processWithdrawalDirect(update prefixUpdate) {
-	// For withdrawals, we need to delete the route from live_routes
+	if err := h.db.DeleteLiveRoute(update.prefix, update.peer); err != nil {
-	if update.originASN > 0 {
+		h.logger.Error("Failed to delete live route",
-		if err := h.db.DeleteLiveRoute(update.prefix, update.originASN, update.peer); err != nil {
+			"prefix", update.prefix,
-			h.logger.Error("Failed to delete live route",
+			"peer", update.peer,
-				"prefix", update.prefix,
+			"error", err,
-				"origin_asn", update.originASN,
+		)
-				"peer", update.peer,
-				"error", err,
-			)
-		}
-	} else {
-		// If no origin ASN, just delete all routes for this prefix from this peer
-		if err := h.db.DeleteLiveRoute(update.prefix, 0, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route (no origin ASN)",
-				"prefix", update.prefix,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
 	}
 }
 
 // processWithdrawal handles removing a route from the live routing table
 // nolint:unused // kept for potential future use
 func (h *PrefixHandler) processWithdrawal(_ *database.Prefix, update prefixUpdate) {
-	// For withdrawals, we need to delete the route from live_routes
+	if err := h.db.DeleteLiveRoute(update.prefix, update.peer); err != nil {
-	// Since we have the origin ASN from the update, we can delete the specific route
+		h.logger.Error("Failed to delete live route",
-	if update.originASN > 0 {
+			"prefix", update.prefix,
-		if err := h.db.DeleteLiveRoute(update.prefix, update.originASN, update.peer); err != nil {
+			"peer", update.peer,
-			h.logger.Error("Failed to delete live route",
+			"error", err,
-				"prefix", update.prefix,
+		)
-				"origin_asn", update.originASN,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
-	} else {
-		// If no origin ASN, just delete all routes for this prefix from this peer
-		if err := h.db.DeleteLiveRoute(update.prefix, 0, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route (no origin ASN)",
-				"prefix", update.prefix,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
 	}
 }
 

internal/routewatch/prefixhandler_test.go

@@ -1,7 +1,10 @@
 package routewatch
 
 import (
+	"context"
 	"errors"
+	"fmt"
+	"slices"
 	"testing"
 	"time"
 
@@ -72,3 +75,263 @@ func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
 		}
 	}
 }
+
+// TestPrefixHandlerAppliesEachPeersLastUpdateInABatch runs updates for one
+// prefix through the prefix handler into a real database. All updates carry the
+// same timestamp, as updates within one second of the feed do. Within a batch,
+// the last update each peer sent for the prefix decides whether that peer has a
+// live route for it, and updates from different peers are all applied. The
+// in-memory route count and prefix distribution must match the route table.
+func TestPrefixHandlerAppliesEachPeersLastUpdateInABatch(t *testing.T) {
+	const (
+		prefix           = "198.51.100.0/24"
+		prefixMaskLength = 24
+		otherPeerIP      = "2001:db8:ffff::2"
+	)
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{
+			Peer:            peer,
+			Path:            ristypes.ASPath{testASNA, testASNB},
+			ParsedTimestamp: ts,
+			Announcements:   []ristypes.RISAnnouncement{{NextHop: peer, Prefixes: []string{prefix}}},
+		}
+	}
+	withdraw := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{Peer: peer, ParsedTimestamp: ts, Withdrawals: []string{prefix}}
+	}
+	announceAndWithdraw := func(peer string) *ristypes.RISMessage {
+		msg := announce(peer)
+		msg.Withdrawals = []string{prefix}
+
+		return msg
+	}
+
+	tests := []struct {
+		name string
+		// Each inner list is handled and then flushed as one batch.
+		batches [][]*ristypes.RISMessage
+		// The peers left with a live route for the prefix, by peer IP.
+		wantPeers []string
+	}{
+		{
+			name:      "two peers announce the prefix",
+			batches:   [][]*ristypes.RISMessage{{announce(testPeerIP), announce(otherPeerIP)}},
+			wantPeers: []string{testPeerIP, otherPeerIP},
+		},
+		{
+			name: "two peers withdraw the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announce(otherPeerIP)},
+				{withdraw(testPeerIP), withdraw(otherPeerIP)},
+			},
+		},
+		{
+			name:    "a peer announces and then withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{{announce(testPeerIP), withdraw(testPeerIP)}},
+		},
+		{
+			name: "a peer withdraws and then re-announces the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{withdraw(testPeerIP), announce(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+		{
+			// RFC 4271 section 4.3: the announcement counts, not the withdrawal.
+			name: "one message announces and withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announceAndWithdraw(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+			if err != nil {
+				t.Fatalf("failed to create database: %v", err)
+			}
+			defer func() { _ = db.Close() }()
+
+			// Built without NewPrefixHandler's flush timer, so the test flushes itself.
+			h := &PrefixHandler{db: db, logger: logger.New()}
+			for _, batch := range tt.batches {
+				for _, msg := range batch {
+					h.HandleMessage(msg)
+				}
+				h.mu.Lock()
+				h.flushBatchLocked()
+				h.mu.Unlock()
+			}
+
+			ctx := context.Background()
+			routes, err := db.GetPrefixDetailsContext(ctx, prefix)
+			if err != nil && !errors.Is(err, database.ErrNoRoute) {
+				t.Fatalf("GetPrefixDetailsContext: %v", err)
+			}
+			var gotPeers []string
+			for _, route := range routes {
+				gotPeers = append(gotPeers, route.PeerIP)
+			}
+			if !slices.Equal(gotPeers, tt.wantPeers) {
+				t.Errorf("peers with a live route for %s = %v, want %v", prefix, gotPeers, tt.wantPeers)
+			}
+
+			var wantDistribution []database.PrefixDistribution
+			if len(tt.wantPeers) > 0 {
+				wantDistribution = []database.PrefixDistribution{{MaskLength: prefixMaskLength, Count: 1}}
+			}
+			assertCounts(t, db, ipv4Version, len(tt.wantPeers), wantDistribution)
+		})
+	}
+}
+
+// TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer runs updates from one peer
+// through the prefix handler into a real database, flushing each message as its
+// own batch, once with IPv4 prefixes and once with IPv6 prefixes. A peer has at
+// most one route for a prefix: a later announcement replaces it whatever its
+// origin AS, and a withdrawal removes it whatever AS the message's AS path ends
+// in. The in-memory route count and prefix distribution must match the route
+// table.
+func TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer(t *testing.T) {
+	families := []struct {
+		ipVersion           int
+		prefix, otherPrefix string
+		// The mask length of both prefixes.
+		maskLength int
+	}{
+		{ipVersion: ipv4Version, prefix: "198.51.100.0/24", otherPrefix: "203.0.113.0/24", maskLength: 24},
+		{ipVersion: ipv6Version, prefix: "2001:db8:1::/48", otherPrefix: "2001:db8:2::/48", maskLength: 48},
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(path ristypes.ASPath, announced string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{
+			Peer:            testPeerIP,
+			Path:            path,
+			ParsedTimestamp: ts,
+			Announcements:   []ristypes.RISAnnouncement{{NextHop: testPeerIP, Prefixes: []string{announced}}},
+		}
+	}
+
+	for _, family := range families {
+		prefix, otherPrefix := family.prefix, family.otherPrefix
+
+		// One message that announces otherPrefix with origin AS testASNB and
+		// withdraws prefix.
+		announceOtherAndWithdraw := announce(ristypes.ASPath{testASNA, testASNB}, otherPrefix)
+		announceOtherAndWithdraw.Withdrawals = []string{prefix}
+
+		tests := []struct {
+			name     string
+			messages []*ristypes.RISMessage
+			// The origin AS of the peer's live route for each prefix. A prefix
+			// left out must have no live route.
+			wantOrigins map[string]int
+		}{
+			{
+				name: "the peer announces the prefix with a new origin AS",
+				messages: []*ristypes.RISMessage{
+					announce(ristypes.ASPath{testASNA}, prefix),
+					announce(ristypes.ASPath{testASNA, testASNB}, prefix),
+				},
+				wantOrigins: map[string]int{prefix: testASNB},
+			},
+			{
+				name: "one message announces a prefix with another origin AS and withdraws the prefix",
+				messages: []*ristypes.RISMessage{
+					announce(ristypes.ASPath{testASNA}, prefix),
+					announceOtherAndWithdraw,
+				},
+				wantOrigins: map[string]int{otherPrefix: testASNB},
+			},
+		}
+
+		for _, tt := range tests {
+			t.Run(fmt.Sprintf("IPv%d %s", family.ipVersion, tt.name), func(t *testing.T) {
+				db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+				if err != nil {
+					t.Fatalf("failed to create database: %v", err)
+				}
+				defer func() { _ = db.Close() }()
+
+				// Built without NewPrefixHandler's flush timer, so the test flushes itself.
+				h := &PrefixHandler{db: db, logger: logger.New()}
+				for _, msg := range tt.messages {
+					h.HandleMessage(msg)
+					h.mu.Lock()
+					h.flushBatchLocked()
+					h.mu.Unlock()
+				}
+
+				for _, p := range []string{prefix, otherPrefix} {
+					routes, err := db.GetPrefixDetailsContext(context.Background(), p)
+					if err != nil && !errors.Is(err, database.ErrNoRoute) {
+						t.Fatalf("GetPrefixDetailsContext(%s): %v", p, err)
+					}
+					var gotOrigins []int
+					for _, route := range routes {
+						gotOrigins = append(gotOrigins, route.OriginASN)
+					}
+					var wantOrigins []int
+					if origin, ok := tt.wantOrigins[p]; ok {
+						wantOrigins = []int{origin}
+					}
+					if !slices.Equal(gotOrigins, wantOrigins) {
+						t.Errorf("origin ASes of the live routes for %s = %v, want %v", p, gotOrigins, wantOrigins)
+					}
+				}
+
+				wantDistribution := []database.PrefixDistribution{
+					{MaskLength: family.maskLength, Count: len(tt.wantOrigins)},
+				}
+				assertCounts(t, db, family.ipVersion, len(tt.wantOrigins), wantDistribution)
+			})
+		}
+	}
+}
+
+// assertCounts checks that the live route count and prefix distribution of one
+// IP version are wantRoutes and wantDistribution, both as held in memory and as
+// read from the route table.
+func assertCounts(t *testing.T, db *database.Database, ipVersion, wantRoutes int,
+	wantDistribution []database.PrefixDistribution) {
+	t.Helper()
+
+	ctx := context.Background()
+	stats, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	tableRoutesV4, tableRoutesV6, err := db.GetLiveRouteCountsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetLiveRouteCountsContext: %v", err)
+	}
+	queryDistributionV4, queryDistributionV6, err := db.GetPrefixDistributionContext(ctx)
+	if err != nil {
+		t.Fatalf("GetPrefixDistributionContext: %v", err)
+	}
+
+	memoryRoutes, tableRoutes := stats.IPv4Routes, tableRoutesV4
+	memoryDistribution, queryDistribution := stats.IPv4PrefixDistribution, queryDistributionV4
+	if ipVersion == ipv6Version {
+		memoryRoutes, tableRoutes = stats.IPv6Routes, tableRoutesV6
+		memoryDistribution, queryDistribution = stats.IPv6PrefixDistribution, queryDistributionV6
+	}
+
+	if memoryRoutes != wantRoutes || tableRoutes != wantRoutes {
+		t.Errorf("IPv%d live routes: %d in memory, %d in the table, want %d",
+			ipVersion, memoryRoutes, tableRoutes, wantRoutes)
+	}
+	if !slices.Equal(memoryDistribution, wantDistribution) ||
+		!slices.Equal(queryDistribution, wantDistribution) {
+		t.Errorf("IPv%d prefix distribution: %v in memory, %v from the table, want %v",
+			ipVersion, memoryDistribution, queryDistribution, wantDistribution)
+	}
+}

internal/server/handlers.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"net"
 	"net/http"
+	"net/netip"
 	"runtime"
 	"sort"
 	"strconv"
@@ -748,8 +749,15 @@ func (s *Server) handlePrefixDetailJSON() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
+		// Look the prefix up in the text form net/netip prints, the form the
-		prefix := prefixParam + "/" + lenParam
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			writeJSONError(w, http.StatusBadRequest, err.Error())
+
+			return
+		}
+		prefix := parsed.Masked().String()
 
 		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {
@@ -912,8 +920,15 @@ func (s *Server) handlePrefixDetail() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
+		// Look the prefix up in the text form net/netip prints, the form the
-		prefix := prefixParam + "/" + lenParam
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			http.Error(w, "Invalid prefix", http.StatusBadRequest)
+
+			return
+		}
+		prefix := parsed.Masked().String()
 
 		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {

internal/server/handlers_test.go

@@ -148,6 +148,58 @@ func TestStatsHandlersAnswerFromMemory(t *testing.T) {
 	}
 }
 
+// TestPrefixDetailHandlersParsePrefix checks that both prefix pages find a
+// stored route whatever valid text form the URL writes its prefix in, and
+// answer 400 for a prefix that does not parse. The tables store the form
+// net/netip prints. Before the fix the URL text was looked up as typed, so
+// "2001:DB8:0:0::/32" was not found although "2001:db8::/32" was stored, and
+// "foo/24" answered 500 (https://git.eeqj.de/sneak/routewatch/issues/71).
+func TestPrefixDetailHandlersParsePrefix(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("database.New: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := db.UpsertLiveRouteBatch([]*database.LiveRoute{
+		{
+			ID: uuid.New(), Prefix: "192.0.2.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: 64500, PeerIP: "198.51.100.1", ASPath: []int{64500}, NextHop: "198.51.100.1",
+			LastUpdated: ts,
+		},
+		{
+			ID: uuid.New(), Prefix: "2001:db8::/32", MaskLength: 32, IPVersion: 6,
+			OriginASN: 64501, PeerIP: "2001:db8::1", ASPath: []int{64501}, NextHop: "2001:db8::1",
+			LastUpdated: ts,
+		},
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	tests := []struct {
+		path string
+		want int
+	}{
+		{"/api/v1/prefix/foo/24", http.StatusBadRequest},
+		{"/prefix/foo/24", http.StatusBadRequest},
+		{"/api/v1/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/api/v1/prefix/192.0.2.1/24", http.StatusOK},
+		{"/prefix/192.0.2.1/24", http.StatusOK},
+	}
+
+	for _, tt := range tests {
+		rec := httptest.NewRecorder()
+		s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, tt.path, nil))
+		if rec.Code != tt.want {
+			t.Errorf("%s: status %d, want %d; body %s", tt.path, rec.Code, tt.want, rec.Body)
+		}
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {

internal/server/server.go

@@ -3,6 +3,8 @@ package server
 
 import (
 	"context"
+	"fmt"
+	"net"
 	"net/http"
 	"strconv"
 	"time"
@@ -53,9 +55,12 @@ func New(db database.Store, streamer *streamer.Streamer, logger *logger.Logger,
 	return s
 }
 
-// Start starts the HTTP server
+// Start listens on the configured port and serves HTTP in the background.
+// It returns an error if it cannot listen, for example because the port is
+// already in use.
 func (s *Server) Start() error {
 	const (
+		network           = "tcp"
 		readHeaderTimeout = 40 * time.Second
 		readTimeout       = 60 * time.Second
 		writeTimeout      = 60 * time.Second
@@ -73,10 +78,15 @@ func (s *Server) Start() error {
 
 	s.logger.Info("Starting HTTP server", "port", s.port, "addr", s.srv.Addr)
 
-	// Start in goroutine but log when actually listening
+	listener, err := net.Listen(network, s.srv.Addr)
+	if err != nil {
+		return fmt.Errorf("starting HTTP server: %w", err)
+	}
+
+	s.logger.Info("HTTP server listening", "addr", listener.Addr().String())
+
 	go func() {
-		s.logger.Info("HTTP server listening", "addr", s.srv.Addr)
+		if err := s.srv.Serve(listener); err != nil && err != http.ErrServerClosed {
-		if err := s.srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
 			s.logger.Error("HTTP server error", "error", err)
 		}
 	}()

internal/server/server_test.go

@@ -0,0 +1,33 @@
+package server
+
+import (
+	"context"
+	"net"
+	"testing"
+
+	"git.eeqj.de/sneak/routewatch/internal/config"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"git.eeqj.de/sneak/routewatch/internal/metrics"
+	"git.eeqj.de/sneak/routewatch/internal/streamer"
+)
+
+// TestStartFailsWhenPortIsHeld holds a port with a listener of its own and
+// checks that Start on that port returns an error. Before the fix Start
+// listened in a goroutine, only logged the failure and returned nil, so the
+// daemon kept running without its web server.
+func TestStartFailsWhenPortIsHeld(t *testing.T) {
+	held, err := net.Listen("tcp", ":0")
+	if err != nil {
+		t.Fatalf("failed to open the test listener: %v", err)
+	}
+	defer func() { _ = held.Close() }()
+
+	port := held.Addr().(*net.TCPAddr).Port
+	log := logger.New()
+	s := New(nil, streamer.New(log, metrics.New()), log, &config.Config{Port: port})
+	defer func() { _ = s.Stop(context.Background()) }()
+
+	if err := s.Start(); err == nil {
+		t.Fatalf("Start returned nil while port %d was held by another listener", port)
+	}
+}